About Helios Artificial Intelligence

Helios Artificial Intelligence, Inc. is dedicated to revolutionizing how agricultural buyers and traders can predict and mitigate supply disruptions due to climate - leveraging advanced data analytics and artificial intelligence.

Need an API key or backtesting data?

Please reach out to [email protected]!

A bit about our Market Watch Data

Description
The Helios Artificial Intelligence, Inc. Agricultural Supply Chain Risk Dataset is a comprehensive collection of climate-driven data meticulously curated and analyzed to assess and mitigate risks across the global agricultural supply chain. This API serves as a foundational resource for stakeholders in the agricultural industry, providing actionable insights and intelligence to inform decision-making processes.

Scope
The API covers 6 (and growing!) of the top traded agricultural commodities (corn, soy, palm-oil, wheat, coffee and cocoa), spanning diverse geographical regions worldwide. It encompasses daily-level data on various climatic factors, including temperature, precipitation, and other relevant parameters, enabling granular analysis of agricultural risks and market dynamics. Check out the Get Available Combinations endpoint to see which countries are covered for each crop!

Source
The primary source of the dataset is climate-driven data obtained from a network of weather stations and environmental monitoring sources globally. Additionally, proprietary data collection tools and algorithms developed by Helios Artificial Intelligence, Inc. are utilized to enhance the richness and accuracy of the dataset. We also incorporate data from the USDA and FAO regarding market share distribution.

Update Frequency
The dataset is updated at 5 am every morning from the previous day's weather to ensure currency and relevance. Daily updates capture the latest climatic data and market trends, providing stakeholders with timely and actionable insights to support their decision-making processes.

Anomaly Detection Algorithms
Proprietary anomaly detection algorithms and clustering techniques are applied to the dataset, enabling the identification of abnormal patterns and trends indicative of potential risks or opportunities. You'll see this manifest on in the data with columns like "too wet", "too dry" etc.

Market Coverage
Extensive research and analysis are conducted to ensure 93% market coverage for each crop globally, enhancing the dataset's utility and relevance for stakeholders across the agricultural supply chain.

Data Collection Process

At Helios Artificial Intelligence, Inc., we have developed a sophisticated data collection methodology that forms the backbone of our anomaly detection algorithms. Our data collection process involves the creation of a comprehensive data mesh covering all land masses globally, comprising approximately 4 million hexagons. These hexagons are grouped into 81 distinct climate profiles, meticulously defined based on various factors including temperature, precipitation, latitude, and elevation characteristics of each location.

Historical Weather Data Analysis

To enrich our dataset, we leverage historical weather reports obtained from nearby weather stations within a 50km radius of the requested locations. The data from the three closest valid weather stations reporting data at the requested time are considered for analysis. These historical records are combined using a weighted mean approach, wherein the closest station holds more weight compared to those farther away. This weighted aggregation ensures a comprehensive representation of the climate conditions surrounding the target location.

Sampling

The sampled data serves as the foundation for our clustering technique, allowing us to identify distinct climate profiles and characterize the unique climate conditions of each location. By leveraging two decades of historical data, we capture long-term climate trends and variations, enabling more accurate and reliable risk assessment and anomaly detection across different regions.

This extensive dataset provides granular insights into the climatic characteristics of diverse geographical regions, facilitating precise analysis and modeling of agricultural risks worldwide. Additionally, our sampling approach ensures broad coverage across all climate profiles, ensuring that our analyses are representative of the global agricultural landscape.

By incorporating this comprehensive sampling approach into our data collection process, we ensure the integrity and reliability of our dataset, empowering our algorithms to deliver actionable insights and mitigate risks effectively across the agricultural supply chain.

Crop Market Coverage

We undertake extensive research and analysis to ensure a robust coverage of 90% of the districts growing each crop for export across the world. This level of coverage is imperative as it enables us to capture the intricacies of market dynamics and climatic conditions influencing agricultural production. By tracking daily weather data on each farming area, we aim to deliver precise and reliable insights into the inherent risks associated with crop cultivation and market fluctuations. Without such comprehensive coverage, our analyses would be susceptible to inaccuracies and limitations.

By integrating this comprehensive approach to crop coverage into our data collection process, we uphold the highest standards of accuracy and reliability, empowering stakeholders to navigate the complexities of the global agricultural landscape with confidence.

Normalization and Adjustment

To maintain consistency and accuracy in our data, the reporting times from the weather stations are normalized to the nearest hour of the day. Additionally, adjustments are made as necessary to account for variations in reporting time zones across different regions. This meticulous normalization and adjustment process ensure that our dataset remains standardized and reliable for subsequent analysis.

Integration with Climate Profiles

The collected weather data is then integrated with our predefined climate profiles, allowing us to assign each location to a specific climate category. This integration enhances the granularity and contextual relevance of our dataset, facilitating more accurate anomaly detection across different regions and climate profiles.

By meticulously curating and integrating climate-driven data from a diverse range of sources, we ensure the robustness and reliability of our dataset, empowering our anomaly detection algorithms to effectively identify and mitigate risks across the global agricultural landscape.

Data Processing and Preprocessing

At Helios Artificial Intelligence, Inc., our anomaly detection algorithms are designed to normalize the risks identified by our proprietary clustering technique. These algorithms aim to yield a standardized level of risk across various categories, making it easily interpretable for humans. We categorize risks into distinct types such as "too hot," "too cold," "too wet," and "too dry," allowing for granular analysis of risk factors affecting agricultural regions worldwide.

Normalization and Standardization of Risk

Our algorithms calculate the percentage of days experiencing a relatively high level of risk over a recent rolling window of time for each risk type, as well as an overall level of risk for the same period. This normalization process enables us to compare and interpret risk levels consistently across different regions and timeframes.

Aggregation at Multiple Levels

Individual Location Level
We assess the degree of risk on an individual location level, providing insights into specific risk factors affecting each area.

Regional Level
We average the risk percentages across growing regions unique to each crop, deriving another level of risk aggregation at the regional level. This regional aggregation enhances our understanding of the broader risk landscape for each crop.

Market-Level Aggregation
Finally, we aggregate our data for the day using a weighted average of each region/country based on their market share and market presence. For example, regions like Argentina may have higher weighting during certain months for crops like corn, while the United States may dominate during different periods. This market-based aggregation ensures that our risk assessments are tailored to the dynamics of the global agricultural market.

By implementing these sophisticated processing and preprocessing techniques, we ensure that our data is not only comprehensive and granular but also actionable for our clients.

How to best utilize our Market Watch data

To cater to a diverse range of client needs and technical capabilities, Helios Artificial Intelligence, Inc. offers several tailored options within our data sets:

Raw Risk Signal Data
Ideal for advanced analytical shops and clients who develop their own custom models. This data provides the unprocessed signals that indicate potential risks or anomalies detected across our global monitoring network. By accessing the raw data, sophisticated users can apply proprietary models and perform complex simulations that align with their specific risk management frameworks. Check out the Daily Risk endpoint to see the live data!

Pre-Weighted Data Columns
For clients who require ready-to-use insights without the need for extensive data manipulation, we offer columns that are pre-weighted by market share and seasonality. This added layer simplifies the analysis, allowing users to directly apply the insights into their business decisions without the complexities of data processing. The weighting factors take into account the relative importance of different markets and the timing of agricultural seasons, enhancing the relevance and accuracy of the risk assessments. Check out the Daily Risk Weighted end point to see the live data!

Both data offerings are designed to empower our clients, whether they are looking for the flexibility of raw data for deep analytical work or the convenience of pre-processed, actionable insights for immediate decision-making. This dual approach ensures that every client can leverage our data in a way that best suits their needs.

Best practices for trading with our data

• How often to trade: The data in this API is best suited for mid-frequency trading (ie. weekly).
• Recommended strategy: Buy on close if the weighted risk score has increased (or is forecasted to increase) by at least 10% (simple difference not the % change). Then sell on the close when the actual price has gone up and starts to come back down (meaning yesterdays weighted risk score is greater than today's). But don't take our word for it! Please reach out to us for backtesting data to see for yourself!
• Market Response Times: Keep in mind that market response times to climatic events tend to vary across crops. Generally we have seen crops that have a large US presence tend to respond within 4 to 7 days. That are grown more internationally can have lag times from three to eight weeks.

FAQs

• Q: What does "x_days" refer to?
  • A: x_days refers to the number of days in the rolling window. For example avg_percent_too_dry_last_x_days would mean the average percent of days that we too dry over the last x days. Currently we are using a 7 days rolling window as it has proved to be most predictive. However please let us know if you would like us to offer other rolling window periods in the future.
• Q: What does it mean when the weighted risk value is null?
  • A: That just means that that crop is out of it's "trading season". Note this is different from a growing season. A "trading season" is the period of highest frequency of trades.